I suppose there are a few things you could look into to make the test results appear 'successful'. On one hand, you could up the number of seconds that are unacceptable for 5% of requests or more in the load settings. You can also 'ignore the first x seconds of the test'. Sometimes, you will get a few requests that are way, way off, and you can possibly filter these out this way.
It also depends on the test scenario -- for instance maybe your site can handle 5 users, but if the Sleep() pauses in the script are really short or nonexistent, it's not a realistic test. A real user might stop to read a page for a few minutes, for instance, so long pages should have long sleep times between them. Any recorded script be accurate already as the Sleep times are the same as the user session that had created the script. If the script was created by hand, though, there is a possibility that you're following one request right after another, which would probably be many more requests per second than 5 users would normally generate.